Stop Counting Tokens
The outlet's thesis cuts to the real economics of the agent era: a task that once cost one prompt now spawns planning, tool calls and retries, so inference demand grows far faster than user counts. It is right that token volume is a vanity metric — buyers pay for working software, not tokens burned.… more
The catch, in our view, is that 'what ships' is devilishly hard to meter: quality, rework and attribution are messy, which is why OpenAI and Anthropic still bill per token. The likely landing spot is hybrid pricing — per-task or per-outcome for commoditized flows, tokens underneath. If agent adoption keeps compounding, outcome-based billing stops being a thought experiment and becomes the margin battle of the next cycle.
Stop counting tokens, start counting what ships
When a developer gives an AI agent a coding task, the agent does more than answer once. It plans, generates code, runs tests, and repeats the process to correct errors. A single task can use 10 to 50 times more tokens than a basic chat prompt. The more useful the agent becomes, the more calls…
Source:
MIT Technology Review
Discussion (0)